See pricing
Why every company now needs to think and operate like a lab team | Josh Woodward (Google Labs, Gemini, AI Studio)
Lenny's Podcast: Product | Career | GrowthAI

Why every company now needs to think and operate like a lab team | Josh Woodward (Google Labs, Gemini, AI Studio)

Josh Woodward explains why the winning AI-era company will not just ship faster, but will build a small team whose job is to live at the frontier and notice when something moves from "almost possible" to real. He also gives unusually concrete signals for early product-market fit, when to kill an idea, and how to design a labs team that does not turn into a toy R&D group.

October 11, 2026 · 62 min listen · 14 min read · Josh Woodward
0:00
–:––

Context

Josh Woodward runs Google Labs, the Gemini app, and AI Studio. His vantage point is unusual: he has to find new AI product opportunities early, test them fast, and decide which ones deserve to scale inside a very large company. This matters to PMs because his core argument is that AI is changing fast enough that every company now needs some version of a labs team, not as a side hobby, but as a way to stay ahead of what the technology just made possible. The episode is especially useful because Josh is concrete about where ideas come from, what early product-market fit actually looks like, and how to know when a project should die.

The Big Idea

In the AI era, companies need a protected team whose job is to live at the frontier, spot when something shifts from "almost possible" to usable, and turn that shift into products before competitors do.

The hard part is not generating more ideas. The hard part is building the culture, rhythms, and judgment to know what to explore, what to scale, and what to kill.

Josh's point is that this is not just a startup skill. Large companies need it too, but they usually fail because they either bury labs work inside the core business or let it drift into disconnected experiments that never change the company.

Key Insights

Every company needs frontier teams

  • Josh agrees with Lenny's premise that every company needs a team whose explicit job is to stay at the frontier of what new models and tools make possible.
  • That team should be trying new tools constantly, understanding new model releases quickly, and spotting shifts before a startup does.
  • The goal is not to distract the whole company with shiny objects. The goal is to let the rest of the company stay focused while a small group learns faster on its behalf.
  • For PMs, this reframes AI work from "everyone should experiment a bit" to "someone must own frontier learning as a real operating function."

Ideas emerge from immersion

Josh is direct that the best ideas rarely come from formal idea-generation rituals like design sprints. They usually show up when curious people are already deep in the work, hacking during breaks, sharing side projects, or reacting to something surprising they just saw. His advice is to stay on alert for what catches your attention, because good ideas tend to appear in unconventional moments, not on a calendar invite.

The implication for PMs is that idea quality depends less on running better brainstorms and more on building an environment where obsessed people are close to new capabilities and free to tinker.

Track the almost possible

  • Google Labs keeps a running view of what is "almost possible".
  • The point is to watch for phase shifts. A capability can look unimportant for months, then cross a threshold in latency, quality, modality, or cost and suddenly support a real product.
  • Josh described this as a change like solid to liquid or liquid to gas. Once the shift happens, the team moves fast.
  • This only works if a team has already done the prep work: forming a view of the future, tracking enabling technologies, and studying user behavior outside the Bay Area bubble.
  • He said Google Labs has a document with 82 predictions about where the future is heading. He fully expects many to be wrong. The value is having a point of view sharp enough to notice when reality starts moving toward one of them.

Early PMF is human

Josh's strongest point on early product-market fit is that it is still more art than science. In the earliest stages, the team is not staring at D7, DAU, or other dashboard metrics. They are watching people react to prototypes.

  • The earliest metric is whether a user visibly leans in.
  • He said the real signal is in people's eyes.
  • If a prototype hits a real nerve, users look energized, curious, and pulled toward it.
  • If the reaction is flat, the team should not talk itself into a win just because the demo looked technically impressive.

His example was hearing an early NotebookLM audio conversation based on British Parliament debates. The source material itself was not inherently exciting. What mattered was that the AI hosts made it genuinely interesting, which created the feeling that something new and valuable had just appeared.

Teams detect failure first

Josh's rule is that the team usually knows a project is failing before the leader admits it. The signals are emotional and practical at the same time: passion starts fading, the team has tried every obvious move, and user reactions stay weak.

He gave a current example from Gemini. A team had a feature they were excited to launch, then went out and watched real reactions. The energy was not there, and the PM recommended not launching it. Josh praised that call publicly. The lesson is that labs cultures need permission to say "this is not good enough" before sunk cost takes over.

AI rewards unlearning and trust

Josh is looking for a different mix of talent than the usual "fast learner" framing.

  • Unlearning rate: how quickly can someone drop an old assumption once reality changes?
  • Explosive endurance: can someone sprint hard when the moment demands it, then sustain that pace over time without burning out?
  • Trust building: can someone work fluidly with both people and AI systems, and make other people want to build with them again?

For PMs, this is important because AI lowers execution costs, which means more things can be tried. That makes judgment, resilience, and collaboration more valuable, not less.

Builders still need specialties

Josh pushes back on the idea that AI is collapsing every role into one generic builder. He does think roles are blurring, and he sees PMs, designers, engineers, strategy people, and others becoming more hands-on. But he does not think specialties disappear.

He uses a major and minor analogy. People may gain builder skills across functions, but they still need a core craft. He especially warns early-career PMs not to let prototyping or vibe coding replace the work of becoming excellent at product management itself.

Mental Models & Frameworks

The almost possible list

  • Keep a living list of capabilities that are close, but not ready.
  • Track what would need to improve for each one to become viable, such as latency, reliability, modality, language support, or cost.
  • When one variable crosses the line, treat it as a trigger to prototype immediately.
  • Use this model when the technology is changing too fast for annual planning to be meaningful.

This is a useful PM framework because it turns vague future-watching into an explicit backlog of opportunities waiting for a technical unlock.

Eyes as early metric

  • Use this in the earliest stage of product discovery, before strong usage data exists.
  • Show rough prototypes to real people and watch for visible excitement, curiosity, and urgency.
  • Ignore the temptation to hide behind quantitative metrics too early.
  • A weak emotional reaction to a clever prototype is a real signal, not a data gap.

Josh's point is not that analytics stop mattering. It is that the first question is whether the product creates any human pull at all.

Problem love over product

Fall in love with the user problem, not the first solution. Josh said it often takes three, four, or five pivots before a team either finds something strong or concludes the idea is dead. A team that is emotionally attached to its first implementation will misread evidence and resist necessary changes.

Use this model whenever a team is exploring a new AI capability. The product form will likely change several times. The user problem should remain the anchor.

Labs lifecycle stages

Josh described labs products as moving through distinct stages:

  • Zero to one: discover whether there is any real pull at all. Optimize for prototypes, user reactions, and conviction.
  • One to 10: turn a promising idea into something that can hold up under broader use. The work becomes more about shaping the product and team.
  • 10 to 100: scale growth, funnels, and repeatability. Different metrics and operating muscle matter here.

This matters because labs teams often fail by applying the wrong metrics or org structure to the wrong stage.

Seasons instead of balance

Josh does not describe lab work as steady-state work-life balance. He describes it as seasons.

  • Some moments are intense because a model release, company event, or product window matters.
  • Other moments should be deliberately looser, with time to hack, explore, and plant the next seeds.
  • Leaders should name the season clearly so people know whether they are in sprint mode or regeneration mode.

This is a practical operating model for AI teams. The mistake is not intensity itself. The mistake is sustained intensity with no rhythm or recovery.

Decision Principles

Principle: Separate labs from core

  • When: setting up an internal labs or frontier team.
  • Why: Josh believes frontier work usually dies when it is tucked neatly inside an existing business unit. It needs enough independence to let weird ideas grow, often with direct senior support.

Principle: Plan on rolling horizons

  • When: planning AI products in a fast-changing environment.
  • Why: Josh said a two-year roadmap no longer fits work like Gemini. Google Labs tends to think in roughly six-month rolling windows and asks what a team can move meaningfully in 50 to 100 days.

Principle: Grow only with proof

  • When: deciding whether to add headcount to a new initiative.
  • Why: Josh has seen teams celebrate growth in people before they have product-market fit. He gave a past example of reaching 30 engineers around an inspiring vision with no real PMF. More people amplified the illusion, not the product.

Trade-offs & Nuance

Independence versus integration

A labs team needs distance from the core business, but not so much distance that the rest of the company dismisses it as a toy shop. Josh says many corporate labs fail not because they lack ideas, but because they never bridge those ideas back into the company. The design challenge is to protect experimentation while still making the work legible and transferable.

Intensity versus sustainability

Josh sees the AI moment as requiring bursts of extreme effort. He also thinks teams burn out if that becomes the only mode. The answer is not avoiding hard pushes. The answer is making the pushes explicit, time-bounded, and followed by periods where people can recover and explore.

Range versus deep craft

AI makes it easier for more people to prototype, code, and simulate adjacent functions. That range is useful. But Josh thinks companies will make a mistake if they interpret this as proof that deep specialties no longer matter. The best builders may be broader than before, but they still have a craft center of gravity.

Common Mistakes

Mistake: Running workshops for ideas

Josh said the best ideas very rarely come from scheduled design sprints or conventional brainstorming sessions. The better approach is to put curious builders near the frontier, let them tinker, and watch closely for the surprising things they bring back.

Mistake: Burying labs in core

A labs team nested too tightly under an existing business line often gets optimized for current priorities instead of future opportunities. Josh argues that weird, early-stage ideas need some protected independence or they will be filtered out before they have a chance to mature.

Mistake: Scaling teams too early

Adding people before there is clear product pull creates false confidence. It can feel like progress because the initiative looks more serious, but it usually increases coordination cost faster than learning speed.

Mistake: Forcing order too soon

Josh is still actively learning how much to isolate versus group teams around themes. His warning is that leaders often rush to make early exploration neat and consolidated. In frontier work, some messiness is useful. If you impose too much structure too early, you can kill the very variation that helps strong ideas emerge.

Practical Application

Build an almost possible tracker

Create a shared document for your product or domain with three columns:

  • capability that is nearly viable
  • blocker that still makes it impractical
  • signal that would tell you the blocker just disappeared

Review it every week. The goal is to stop treating technical progress as background noise and start treating it as a trigger for product discovery.

Demo prototypes for eye tests

Before the next major roadmap commitment, show rough prototypes to real users or internal proxies and watch their reactions live.

  • Do not ask only for verbal feedback.
  • Note whether people lean in, ask follow-up questions, or try to use the product immediately.
  • Treat flat energy as evidence, even if the prototype seems clever.

This gives a PM an early signal before they invest in building the polished version of something nobody truly wants.

Plan in 100-day windows

For any AI-heavy initiative, replace long deterministic roadmaps with a rolling plan that answers one question: what meaningful milestone can this team hit in the next 50 to 100 days?

That milestone could be a prototype, a user behavior insight, a narrower problem definition, or evidence strong enough to justify more investment. The point is to shorten the distance between learning and replanning.

Write a labs charter

If you are setting up a frontier team, write a one-page charter that answers four things:

  • what user or strategic frontier the team watches
  • how independent the team is from current business units
  • whether success means graduating ideas into existing products or creating new product lines
  • how and when the team hands work back to the rest of the company

Josh's warning is that many labs fail because none of this is explicit.

Screen for unlearning and trust

When hiring or staffing exploratory teams, look beyond raw intelligence or resume prestige.

  • Ask for an example of a belief the person changed quickly after seeing new evidence.
  • Ask how they handle ambiguous work where the path is unclear.
  • Ask for examples of how they built trust across disciplines.
  • Give extra credit if the person naturally shows you things they built without being prompted.

These are closer to the traits Josh thinks matter in frontier product work.

Questions to Consider

  • What has become newly possible in your product area in the last 90 days, and who on your team is explicitly responsible for noticing that shift before competitors do?
  • Which current roadmap items look strong in a planning document but fail the "users' eyes light up" test when shown as a prototype?
  • Where is your organization asking a frontier team to behave like a normal business unit too early, and what promising weirdness might that be killing?
  • Which project on your team is being kept alive by effort already spent rather than by visible user pull or team conviction?
  • If every role on your team is becoming more hands-on with AI tools, what core craft should each function still treat as its major, not just its minor?

Bottom Line

Josh Woodward's core lesson is that AI-era product teams need a labs mindset: stay close to the frontier, notice when "almost possible" becomes real, and move fast with small teams and live prototypes. The companies that win will not be the ones with the prettiest AI roadmap. They will be the ones with the clearest judgment about what users truly want, what to kill early, and how to protect weird ideas long enough to matter.

Tools & Products

Tool / ProductWhat it doesWhy it was mentioned
Google LabsGoogle's internal frontier product groupJosh uses it as the model for how companies should explore new AI opportunities without distracting the whole organization.
Gemini appGoogle's consumer AI assistant productJosh contrasts its global scale with earlier-stage labs work to show that both still require a zero-to-one mindset in parts of the product.
AI StudioGoogle's tool for building and experimenting with AIIt is part of the set of products Josh oversees while thinking about how frontier ideas become usable products.
NotebookLMAI research and synthesis productJosh used the first time he heard an AI-generated NotebookLM conversation as an example of a prototype that instantly signaled new product potential.
FlowGoogle's AI creative productJosh cited early Flow and WISC experiments as examples of seeing a new level of creative control become possible.
Google BeamAI-powered 3D communication systemJosh mentioned it to show that Google Labs is not limited to pure software and still explores adjacent hardware experiences.

People to Follow

Josh Woodward

Josh Woodward leads Google Labs, the Gemini app, and AI Studio. He is worth following because he is thinking in public about how large companies can preserve startup-style experimentation, how AI changes product craft, and how to build cultures that know when to keep going and when to quit.

Notable Quotes

"You can't microwave good ideas." (Josh Woodward)

"When you're showing people early prototypes, what I'm always doing in those is you're looking at people's eyes. And that is the metric." (Josh Woodward)

"The team usually knows before the leader knows." (Josh Woodward)

"I'm actually looking a lot right now for people. What's their unlearning rate?" (Josh Woodward)

AI PM course

Everyone hears the same episodes.
Few can do what they describe.

Start for free